regional analysis of user interactions on social media in ... · regional analysis of user...
TRANSCRIPT
Regional Analysis of User Interactions on Social Media inTimes of Disaster
Takeshi SakakiThe University of [email protected]
tokyo.ac.jp
Fujio ToriumiThe University of Tokyo
Kosuke ShinodaRiken
Kazuhiro KazamaWakayama University
Satoshi KuriharaOsaka University
Itsuki NodaNational Institute of Advanced
Industrial Science andTechnology
Yutaka MatsuoThe University of [email protected]
tokyo.ac.jp
ABSTRACTSocial media attract attention for sharing information, es-pecially Twitter, which is now being used in times of dis-asters. In this paper, we perform regional analysis of userinteractions on Twittter during the Great East Japan Earth-quake and arrived at the following two conclusions:Peoplediffused much more information after the earthquake, espe-cially in the heavily-damaged areas; People communicatedwith nearby users but diffused information posted by dis-tant users. We conclude that social media users changedtheir behavior to widely diffuse information.
Categories and Subject DescriptorsJ.4 [Computer Applications]: Social and behavioral sci-ences
General TermsHuman Factors, Measurement
KeywordsTwitter, online social network, information diffusion, disas-ter situation, earthquake,regional analysis
1. INTRODUCTIONDuring a disaster, collecting information is important to
save lives. Victims require information about shelters orespecially dangerous points. Furthermore, rescuers requireinformation such as victim locations or the availability ofsupplies.Social media attract attention for their information shar-
ing capabilities, especially Twitter, which is one hugely pop-ular social medium that is used during disasters [1],[2]. By
Copyright is held by the author/owner(s).WWW 2013 Companion, May 13–17, 2013, Rio de Janeiro, Brazil.ACM 978-1-4503-2038-2/13/05.
analyzing interaction behaviors on Twitter, we can estimatehow people use social media during crises.
In the previous research, we revealed that the use of Twit-ter was different from area to area[3]. In this paper, we focuson regional differences in user interactions of Twitter duringthe Great East Japan Earthquake that occurred at 14:46 onMarch 11, 2011.
We prepared a dataset of tweets by crawling those postedby 1.3 million Japanese Twitter users fromMarch 7 to March23. We collected 362, 435, 649 tweets, which were posted by2, 711, 473 users.
2. REGIONAL ANALYSIS OF USER INTER-ACTIONS
In case of the Great East Japan Earthquake, the degree ofdamage caused by it varied depending on the area. We as-sume that the variety affected the user interaction on Twit-ter. Therefore, we performed a regional analysis of userinteractions on it around the earthquake.
First, we extracted location information of Twitter userand selected target areas for analysis. Second, we analyzedthe changes in the number of replies and retweets. Third,we compared them based on the distance between the usersmaking interactions.
2.1 Regional Information on TwitterWe estimated user locations from their Twitter profiles
and analyzed the user interactions by prefecture based onit. However, showing all analysis results is difficult due tospace limitations. We chose five representative prefecturesbased on the degree of damage(Table 1).
We compared the following two types of user interaction:
• Reply : messages for a specific user. We treat a replyas communication interaction.
• Retweet : messages to diffuse a tweet to followers. Wetreat a retweet as diffusion interaction.
235
(a) Replies by selected prefectures
(b) Retweets by selected prefectures
Figure 1: Replies and retweets by selected prefec-ture
2.2 Comparison of Number of User Interac-tion
Figure 1 represents the ratios of the reply and retweetcounts posted by users who apparently live in different pre-fectures. We used all the replies and retweets posted onMarch 7 for normalization.First, there was little change in the reply graphs of all the
prefectures thorough the analysis period compared to thoseof the retweets, which maintained higher levels at the end ofthe analysis period than at the beginning. People diffusedmore information on Twitter than before the earthquake.Second, the characteristics of the retweet graphs differ
depending on the area. Shortly after the earthquake, inthe less-damaged areas and Fukushima, the retweet ratioincreased determinately; in the heavily-damaged areas, theretweet ratio only slightly increased. One week after theearthquake, the retweet ratio in the heavily-damaged areasmaintained higher levels. After a nuclear accident got worse,the retweet ratio in Fukushima increased much more.
2.3 Comparison of User Interaction by Dis-tance
Next, we compared two types of user interactions by theirdistance in the real world with interaction distance, whichwe define as the following distance between two users who
Table 1: Prefecture circumstancesPrefecture Damage Population OthersIwate catastrophic small -Miyagi catastrophic small -Fukushima heavy medium nuclear power
plant destructionTokyo light very large -Osaka - large -
Figure 2: Interaction distance of reply and retweets
perform one user interaction:
dij = distance (useri, userj)
We used the location information of prefectural governmentoffices for the user location information for simplicity. Forexample, useri is in Tokyo and userj is in Osaka; useriposted the tweet : @userj Are you all right?, dij is thedistance between the Tokyo and Osaka government offices.If dij is smaller, it is likely that useri and userj know eachother, and vice versa.
We used d̄ for one day to compare replies and retweets.Fig. 2 represents the changes in their d̄, which have similarlevels: about 260 km before the earthquake. However, theyhave clearly different characteristics. d̄replies decreased af-ter the earthquake and maintained lower levels for severaldays. But shortly after the earthquake, d̄retweets maintainedhigher levels during post-quake chaos week.
3. CONCLUSIONWe analyzed 360 million tweets that were posted before
and after the Great East Japan Earthquake that occurred onMarch 11th 2011 to elucidate how people interact on Twitterduring disasters. We arrived at the following conclusions:
1. People diffused much more information after the earth-quake, especially in the heavily-damaged areas.
2. People communicated with nearby users but diffusedinformation posted by distant users
.From our results, we conclude that social media users changedtheir behavior and reasons to autonomously use social me-dia after serious events. In the future, we must observe andanalyze the changes in social media to ascertain how suchdisasters affect social media over long periods.
4. ACKNOWLEDGMENTSWe also thank Miscrosoft Research Asia for providing nec-
essary financial assitance, and Genta Kaneyama (CookpadInc.) for assistance in collecting data from Twitter.
5. REFERENCES[1] T. Heverin and L. Zach. Microblogging for Crisis
Communication: Examination of Twitter Use inResponse to a 2009 Violent Crisis in Seattle-Tacoma,Washington Area. In Proceedings of ISCRAM 2010,2010.
[2] M. Miyabe, E. Aramaki, and A. Miura. Use trendanalysis of twitter after the great east japanearthquake. In Proceedings of SIG-DPS/GN2011-DPS-148/2011-GN-81/2011-EIP-53, 2011.
[3] T. Sakaki, F. Toriumi, and Y. Matsuo. Tweet trendanalysis in an emergency situation. In Proceedings ofSWID’11, pages 3:1–3:8. ACM, 2011.
236